AMD Radeon Pro 575X vs NVIDIA P104-100 Comparison

FAQ

Q: Which GPU has the higher average benchmark score?

A: The AMD Radeon Pro 575X records an average benchmark score of 39,116, which places it at the 82nd percentile among all GPUs. The NVIDIA P104-100 records a lower average benchmark score of 32,982, placing it at the 77th percentile.

Q: How do the two cards compare in Geekbench OpenCL performance?

A: The NVIDIA P104-100 scores 52,368 in Geekbench OpenCL, which is 33.6% higher than the AMD Radeon Pro 575X’s score of 34,773. This is the largest performance gap between the two cards in any recorded test.

Q: What about Vulkan performance?

A: The NVIDIA P104-100 also leads in Geekbench Vulkan with a score of 45,165, while the AMD Radeon Pro 575X scores 37,919. That gives the NVIDIA card a 16% advantage in this API test.

Q: Do both cards have the same memory capacity?

A: Yes, both GPUs ship with 4 GB of memory. However, the AMD card uses GDDR5 memory with a 256-bit bus and 217.0 GB/s of bandwidth, while the NVIDIA card uses GDDR5X memory with a 256-bit bus and 320.3 GB/s of bandwidth.

Q: What are the nearest rivals to each card according to the database?

A: The AMD Radeon Pro 575X’s closest competitors are the AMD Radeon Pro 580X (average score 38,706, a 1.1% difference), the NVIDIA GeForce MX570 A (38,691, 1.1% difference), the AMD Radeon Pro 575 (39,555, a 1.1% difference in the opposite direction), and the NVIDIA RTX A500 Mobile (39,568, also a 1.1% difference). The NVIDIA P104-100 sits near the NVIDIA T600 Mobile (32,849, 0.4% difference), the NVIDIA T550 Mobile (33,161, 0.5% difference), the NVIDIA GeForce RTX 3050 Mobile (33,170, 0.6% difference), and the AMD Radeon Pro 570 (33,207, 0.7% difference).

Q: Which card has a higher pixel fill rate?

A: The NVIDIA P104-100 has a pixel rate of 110.9 GPixel/s, which is substantially higher than the AMD Radeon Pro 575X’s 35.07 GPixel/s. The NVIDIA card also has 64 ROPs compared to 32 on the AMD card.

Architecture Differences

The AMD Radeon Pro 575X is built on the Ellesmere chip using GCN 4.0 architecture, fabricated on a 14 nm process at GlobalFoundries. It contains 5,700 million transistors on a 232 mm² die, yielding a transistor density of 24.6 million transistors per square millimeter. The NVIDIA P104-100 uses the GP104 chip with Pascal architecture, manufactured by TSMC on a 16 nm process. This chip contains 7,200 million transistors on a 314 mm² die, for a density of 22.9 million transistors per square millimeter. The AMD chip is smaller and denser, while the NVIDIA chip packs more total transistors.

The compute configurations differ considerably. The AMD Radeon Pro 575X carries 2,048 shading units, 128 texture mapping units, and 32 ROPs. The NVIDIA P104-100 has 1,920 shading units, 120 TMUs, and 64 ROPs. Although NVIDIA uses fewer shaders and TMUs, its ROP count is exactly double that of the AMD card. The ROP advantage directly explains the large pixel rate gap: 110.9 GPixel/s versus 35.07 GPixel/s.

Memory subsystems also diverge. The AMD card uses 4 GB of GDDR5 on a 256-bit bus, with a memory clock of 1,695 MHz and 6.8 Gbps effective speed, producing 217.0 GB/s of bandwidth. The NVIDIA card uses 4 GB of GDDR5X on the same 256-bit bus, but with a memory clock of 1,251 MHz and 10 Gbps effective speed, reaching 320.3 GB/s. The GDDR5X memory gives NVIDIA a 47% bandwidth advantage, which matters for bandwidth-hungry workloads.

FP32 throughput favors NVIDIA. The P104-100 delivers 6.655 TFLOPS, while the Radeon Pro 575X delivers 4.489 TFLOPS. FP16 performance shows a stark contrast: the AMD card achieves 4.489 TFLOPS at a 1:1 ratio with FP32, while the NVIDIA card manages only 104.0 GFLOPS at a 1:64 ratio. This means the AMD card is far more capable in FP16 compute, whereas the NVIDIA card treats FP16 as a minor feature.

Power and physical design differ as well. The AMD Radeon Pro 575X has a TDP of 150 W, uses no power connectors, and is listed as an integrated graphics package (IGP) with display outputs described as portable device dependent. The NVIDIA P104-100 lists no TDP in the database, but its suggested power supply is 200 W. It is a dual-slot card, 267 mm long (10.5 inches), requires a single 8-pin power connector, and has no display outputs at all. This reflects its mining GPU lineage.

API support shows minor differences. Both support DirectX 12 and OpenGL 4.6. The AMD card supports Vulkan 1.3, while the NVIDIA card supports Vulkan 1.4. In DirectX feature level terms, the AMD card is rated 12_0 and the NVIDIA card is rated 12_1, granting NVIDIA slightly more advanced DirectX feature support.

The bus interfaces also differ. The AMD card uses PCIe 3.0 x16, a standard full-bandwidth interface. The NVIDIA P104-100 uses PCIe 1.0 x4, a much older and narrower interface. Despite this limitation, the NVIDIA card still wins in raw compute tests, which suggests that memory bandwidth and compute throughput dominate in the recorded benchmarks.

Production status is end-of-life for both cards. The AMD Radeon Pro 575X was released in March 2019, while the NVIDIA P104-100 was released in December 2017.

Head-to-Head Benchmarks

The database records two head-to-head benchmark comparisons between these GPUs, and the NVIDIA P104-100 wins both.

In Geekbench OpenCL, the NVIDIA P104-100 scores 52,368 against the AMD Radeon Pro 575X’s 34,773. This is a 33.6% lead for NVIDIA. The magnitude of this gap is notable given that the AMD card holds a higher overall average benchmark score across all recorded tests. The OpenCL result likely reflects the NVIDIA card’s higher FP32 throughput (6.655 TFLOPS versus 4.489 TFLOPS) and its much larger memory bandwidth (320.3 GB/s versus 217.0 GB/s). Compute-heavy OpenCL workloads that scale with raw shader throughput and memory bandwidth would favor the NVIDIA card.

In Geekbench Vulkan, the NVIDIA P104-100 scores 45,165, while the AMD Radeon Pro 575X scores 37,919. The NVIDIA card leads by 16%. This is a smaller margin than in OpenCL, but still a clear victory. The AMD card’s Vulkan 1.3 support versus NVIDIA’s Vulkan 1.4 does not close the gap. The NVIDIA card’s higher pixel rate and texture rate (110.9 GPixel/s and 208.0 GTexel/s versus 35.07 GPixel/s and 140.3 GTexel/s) may contribute to its advantage in graphics-oriented Vulkan workloads.

The AMD Radeon Pro 575X has no head-to-head wins in the database. It trails in both recorded tests. However, its average benchmark score across all tests (39,116) is higher than the NVIDIA card’s average (32,982). This discrepancy comes from the fact that the AMD card has three recorded benchmark scores (Geekbench Metal 44,655, Geekbench OpenCL 34,773, Geekbench Vulkan 37,919), while the NVIDIA card has three scores as well (3DMark Steel Nomad DX12 1,413, Geekbench OpenCL 52,368, Geekbench Vulkan 45,165). The NVIDIA card’s very low 3DMark result pulls its average down, even though it dominates in the compute-oriented Geekbench tests.

The Geekbench Metal result for the AMD card is particularly strong: 44,655, which is higher than its Vulkan score and far higher than its OpenCL score. The database does not list a Metal test for the NVIDIA card, so no direct comparison is possible. Still, this Metal score helps explain why the AMD card’s average benchmark score sits at the 82nd percentile, above the NVIDIA card’s 77th percentile.

The nearest rival data reinforces the positioning. The AMD Radeon Pro 575X sits within 1.1% of the AMD Radeon Pro 580X, NVIDIA GeForce MX570 A, AMD Radeon Pro 575, and NVIDIA RTX A500 Mobile. Its average score of 39,116 is slightly above its nearest rival on the low side (38,691, the MX570 A) and slightly below its nearest rival on the high side (39,568, the RTX A500 Mobile). The NVIDIA P104-100, by contrast, sits within 0.7% of the NVIDIA T600 Mobile, NVIDIA T550 Mobile, NVIDIA GeForce RTX 3050 Mobile, and AMD Radeon Pro 570. Its average score of 32,982 is only 0.4% above the T600 Mobile (32,849) and 0.7% below the Radeon Pro 570 (33,207).

The overall picture is that the NVIDIA P104-100 is stronger in pure compute and graphics throughput per the head-to-head tests, while the AMD Radeon Pro 575X has a better overall average benchmark score, driven by its strong Metal showing and the NVIDIA card’s weak 3DMark Steel Nomad result.

Specification Differences

The two cards differ across nearly every specification category.

The manufacturing process differs: AMD uses a 14 nm node at GlobalFoundries, while NVIDIA uses a 16 nm node at TSMC. The NVIDIA chip has more transistors (7,200 million versus 5,700 million) and a larger die (314 mm² versus 232 mm²). Transistor density is slightly higher on the AMD chip (24.6 million per mm² versus 22.9 million per mm²).

Core configuration: AMD has 2,048 shading units, 128 TMUs, and 32 ROPs. NVIDIA has 1,920 shading units, 120 TMUs, and 64 ROPs.

Clock speeds: The AMD card lists no base or boost clock in the database, only a memory clock of 1,695 MHz (6.8 Gbps effective). The NVIDIA card lists a base clock of 1,607 MHz and a boost clock of 1,733 MHz, with a memory clock of 1,251 MHz (10 Gbps effective).

Memory: Both have 4 GB, but AMD uses GDDR5 with a 256-bit bus and 217.0 GB/s bandwidth, while NVIDIA uses GDDR5X with a 256-bit bus and 320.3 GB/s bandwidth.

Compute rates: AMD achieves 35.07 GPixel/s pixel rate, 140.3 GTexel/s texture rate, 4.489 TFLOPS FP32, and 4.489 TFLOPS FP16 (1:1). NVIDIA achieves 110.9 GPixel/s, 208.0 GTexel/s, 6.655 TFLOPS FP32, and 104.0 GFLOPS FP16 (1:64).

Power and physical: AMD has a TDP of 150 W, an IGP slot width, no power connectors, and no listed PSU requirement. NVIDIA has no TDP listed, a dual-slot width, one 8-pin power connector, a suggested PSU of 200 W, and a length of 267 mm (10.5 inches).

Interface and outputs: AMD uses PCIe 3.0 x16 and has display outputs listed as portable device dependent. NVIDIA uses PCIe 1.0 x4 and has no display outputs.

API support: Both support DirectX 12 and OpenGL 4.6. AMD supports Vulkan 1.3 and DirectX feature level 12_0. NVIDIA supports Vulkan 1.4 and DirectX feature level 12_1.

Release dates: AMD launched in March 2019, NVIDIA in December 2017. Both are end-of-life.

The Verdict

The data describes two very different products. The AMD Radeon Pro 575X is a mobile-oriented, integrated graphics package with 150 W TDP, no external power connectors, and display outputs tied to the host device. It belongs to the Radeon Pro Mac series and is built around GCN 4.0 with strong FP16 support and a solid Metal score of 44,655. Its average benchmark score of 39,116 places it at the 82nd percentile, above the NVIDIA P104-100’s 77th percentile.

The NVIDIA P104-100 is a mining-oriented card from 2017, built on Pascal, with no display outputs, a PCIe 1.0 x4 interface, and a dual-slot cooler requiring an 8-pin connector. It has no listed TDP, but its suggested PSU is 200 W. Its strengths are raw throughput: higher FP32 (6.655 TFLOPS), higher memory bandwidth (320.3 GB/s), double the ROPs (64 versus 32), and much higher pixel and texture rates. In the head-to-head tests, it beats the AMD card by 33.6% in OpenCL and by 16% in Vulkan.

Who should pick which card depends on the workload and platform. For users who need a GPU with display outputs and whose system expects an integrated-style mobile GPU, the AMD Radeon Pro 575X is the only sensible choice, as the NVIDIA card cannot drive a display at all. The AMD card also offers far better FP16 performance (4.489 TFLOPS versus 104.0 GFLOPS) and a higher average benchmark score overall.

For users who prioritize raw OpenCL or Vulkan compute performance and do not need display output, the NVIDIA P104-100 is the stronger card. Its 33.6% OpenCL lead and 16% Vulkan lead are decisive, and its higher FP32 throughput and memory bandwidth align with compute-heavy workloads. However, its PCIe 1.0 x4 interface could bottleneck data transfer in some systems, and its low 3DMark Steel Nomad score of 1,413 suggests it is not a general-purpose graphics winner.

The verdict from the data is clear: the AMD Radeon Pro 575X is the better-rounded card for systems requiring display output and balanced performance, while the NVIDIA P104-100 is a compute-focused accelerator that wins the head-to-head tests but lacks the features needed for a typical desktop or mobile graphics role.